From Interaction Data to Plan Libraries: A Clustering Approach

Abstract

Plan libraries are the most important knowledge source of many plan recognition systems. The plan decompositions they contain provide information about how a plan has to be executed to actually achieve its associated goals and be recognized by the system. This paper presents an approach to the automatic acquisition of plan decompositions from sample action sequences. In particular a clustering algorithm is introduced that allows groups of "similar" sequences to be discovered and used for the generation of plan libraries. Empirical tests indicate that these libraries can indeed be successfully used for plan recognition purposes. 1 Introduction Plan libraries are the most important knowledge source of many plan recognition systems. They not only contain all possible types of plans (or goals) to be recognized by such a system---thus delimiting the search space of possible plan hypotheses---, but also represent the details of how these plans have to be executed to actually a...